Hacker News Reader

Comment by JoshGlazebrook | Hacker News Reader

That last part shouldn't be hard with the overpriced bandwidth costs AWS charges.

Curious as to why you say that. Do you mean in comparison to other cloud infrastructure providers or in comparison or rolling your own infrastructure?

Comparison: EC2 charges 9 cents per GB, linode charges 2 cents and starts with a much larger (2TB vs 1GB) free quota. Rackspace charges 12 cents, Azure about 9 cents too.

Those figures suggest that linode seriously oversubscribes their service and expects people not to really use it. EC2 and Azure are in the right cost range. And Rackspace's customer service is expensive.

At the exchange levels, you don't pay for traffic... you pay a subscription membership for the IX and you pay for your hardware; you might have some contracts for peering and transit that could cost money, but no where is it done based on data flowing through the network... It's all just to extract money based on usage with the rationale that people who use more can pay more, but it's not like they get a higher bill when they have more traffic, therefore, you can't say that a provider is selling their BW to cheap; maybe they already paid for all their fixed costs via their compute pricing....

Pure B.S. Cloud providers over charge for bandwidth because they can, they treat it like a luxury cost like RAM where if you need more bandwidth you can usually afford to pay for it. It's not the "right" cost, it's the price AWS set which Azure copied.

Bandwidth is dirt cheap outside of the Cloud, e.g. I'm getting 30 TB of bandwidth as part of my 64GB RAM / 500GB SSD / Quad-Core i7 Skylake for €39 /mo (https://www.hetzner.de/us/hosting/produkte_rootserver/ex41ss...). Which roughly equates to €0.0009 /GB that also includes the cost of hosting entire server with resources that would cost an order of magnitude more on AWS/Azure.

> I'm getting 30 TB of bandwidth as part of my 64GB RAM / 500GB SSD / Quad-Core i7 Skylake for €39 /mo

Yes, you're getting cheap pricing from an oversubscribed line.

Or not, just ran a speed test: getting 160% more bandwidth on Hetzner than my EC2 instance.

One data point is clearly indicative enough to be conclusive.

or +1 data point more than all other baseless responses, but apparently still not enough effort to refute BS lawn-chair factoids.

Oversubscribed means there's a chance you won't receive the rated speed. Just because tonight's results were good, doesn't guarantee you'll always achieve that.

> Bandwidth is dirt cheap outside of the Cloud

Enterprise level network equipment and infrastructure are extremely expensive. Unusually cheap BW rate usually means cheap equipment or over-subscription or not enough qualified support personnel. And in some cases under-selling to get penetration to a market.

The reason Google Fiber can sell cheaply because (AFAIK), in almost every town or cities that they deployed their network they negotiated special deals with municipalities or equivalent entity to get free access to existing infrastructure or get special deals. There is a reason why Google Fibre is not everywhere or they are not pushing it very aggressively. Because building networks are freaking expensive even for google.

BW may not have any value but building the network and maintaining it to serve you BW is expensive. Your BW cost is a reflection of the cost of your network.

Disclaimer: I own an ISP.

> There is a reason why Google Fibre is not everywhere or they are not pushing it very aggressively. Because building networks are freaking expensive even for google.

I am not sure if that is entirely true. At least, Google owns the fibers between any 2 google data centers. It is probably only the last-mile that needs municipality support.

I was specifically referring to last mile fibers in reply to OP implying that outside the cloud business internet is cheap. In this context, I was not talking about cloud business and connectivity cost within data centers (which are also not cheap btw).

Nope, they all just overcharge like crazy. Bandwidth from any traditional data center will cost you (at most) $0.0015 per GB. That's about 60x cheaper than what AWS costs.

AWS only kind of makes sense until your bill starts approaching that of a full time engineer's salary. Then you can slash costs like crazy by rolling your own infrastructure (at higher risk of downtime). Or pull a Netflix and negotiate wholesale prices.

Is that really true? In my experience, "real" data centers always charge by bandwidth, not traffic. You can get 10mbps, or 100, or 1000, but it makes no difference how much data you push. (95th percentile is also common, if you don't commit to a full link.) I've always thought of data quotas as a "consumer" level service. After all, the wires that make up data links don't care about the amount of traffic.

Yep, that's technically more accurate. I should have added the qualifier that, "for those centers that do charge by traffic this is roughly what they'll charge".

I think the other clouds are actually rather undersubscribed since most people run their stuff on AWS.

At least you can leverage Digital Ocean, Linode, etc to run varnish caches to offload bandwidth for things like static assets while keeping your main workloads on AWS. You can build this sorta thing in a couple days I'd say.

Those figures suggest that linode seriously overcharges for bandwidth.

If it's overpriced, presumably customers will shop elsewhere or avoid cloud services altogether?

When overpricing happens, it's usually because the customers had nowhere else to go in the first place.

I guess the amount you can overprice is proportional to the value you add. But then is it really over pricing?

It's significantly overpriced, though most services don't demand enough bandwidth to make it a big enough cost disadvantage to turn them away from eg AWS. That isn't the case for all services, but it is the case for most. If you're big enough to run up a large bandwidth bill on AWS, more often than not you can afford it; if you're not big, you won't typically be running up that kind of bandwidth bill. There are cases that are the exception of course (high traffic services with poor business models, video heavy sites with weak monetization, etc).

I've seen good arguments that the cloud services keep bandwidth costs high on purpose to restrict the types of services that attempt to use (or abuse) their networks. I haven't seen any other great arguments that explain the massive gap between a typical large dedicated host and eg Azure / Google / AWS (all of which can command far cheaper bandwidth costs).

I can speak for Google, not AWS or Azure. Google's network is smart in several ways:

- If you are trying to send a packet from one Google AZ to another, it will traverse Google network only, never hitting the public web.. without a VPN/VPC setup

- If you are trying to reach a service on Google Network, you hit the Google frontend at one of the many Points of Presence around the world, from which point it's a straight path to nearest DC

- If you are trying to reach a customer from your VMs on Google, your packet will be taken as close to your customer as possible on Google backbone

For "cheaper egress" there are CDN providers, and both AWS and Google have their own CDNs as well.

Reply on news.ycombinator.com